Making AI more accessible for people with speech differences

Contact: Erin Flynn
May 31, 2023
Hope Gerlach-Houck sits next to her computer at her desk.
Dr. Hope Gerlach-Houck is on a team working to create voice-activated technology that is accessible for people who stutter.

KALAMAZOO, Mich.—Siri and Alexa can order groceries, turn on lights and appliances and even discover obscure facts to entertain our children. But there's one thing our AI helpers haven't mastered: patience. And it's a huge problem—especially for people who stutter and those with other speech differences.

"We're all misunderstood by Siri and Alexa—it happens incessantly to everyone—but it especially happens to people who stutter," says Dr. Hope Gerlach-Houck, assistant professor of speech, language and hearing sciences and co-principal investigator on a National Science Foundation project aimed at creating voice-activated technology that is accessible to people who stutter.

Right now, inaccessibility of voice-activated technology is robbing people who stutter of that. And the longer this persists, the more inequity is going to widen."

Gerlach-Houck is part of a multidisciplinary team including researchers from ¾ÅÒ»Â鶹ÖÆƬ³§ State University who were awarded an NSF Convergence Accelerator Grant to shift development of the technology into high gear. It's a tall order: Within a year, the team is expected to research, conceptualize and prepare a product for market.

Female professional working on laptop and talking into a speaker.
"We're all misunderstood by Siri and Alexa—it happens incessantly to everyone—but it especially happens to people who stutter," says Gerlach-Houck. "This is about being understood, being seen as a human being and being known for who you are."

The urgency reflects the increasing integration of AI in all aspects of our lives—from health care to employment systems. Several large companies, for instance, now use AI-driven interviews to prescreen applicants. Gerlach-Houck's prior research has demonstrated that, compared to people who don't stutter, stutterers are more likely to be unemployed, underemployed and earn less money for the same work.

"We know that people who stutter are already disadvantaged in the workplace; there's inequity in quality of life. As this type of technology is increasingly adopted, those gaps are going to widen," says Gerlach-Houck. "If AI can't understand what you're saying, you're not going to have a fair shake."

Gerlach-Houck's role on the NSF project team focuses on the stigma and potential biases embedded in the technology. The team is currently surveying a diverse pool of stutterers to identify what type of voice-activated technology they currently use and where it falls short.

"We want to get a good feel for how much they're using it, why they are not using it and what barriers and facilitators they have encountered," she says, putting emphasis not only on accessibility but also on user experiences. "Even if AI understands what people who stutter are saying, AI still has built-in biases related to how people look, how they speak and if they stutter. Inaccessible and biased AI is a threat to human rights."

Other groups have attempted similar projects, but she believes her team has the key to breaking down previous barriers because of the inclusive nature of their project. A person who stutters and the director of a nonprofit stuttering organization are part of their core research team.

"We are approaching the project with a mindset that improved technology is not enough, and that true change will require social and political action in conjunction with the stuttering community," says Gerlach-Houck.

Her team is in the process of collecting and inputting a variety of existing and live speech samples from the stuttering community. If their technology is successful, it could open the door to increased accessibility for many people.

"If we have technology that's more accessible for people who stutter, it's going to have trickle down effects to other communities, including people whose communication varies from the norm or from the data sets that voice-activated technology systems are trained on, which are typically pretty homogenous in terms of dialect."

In addition to creating the technology, the team coordinated an international conference on voice-activated AI, bringing together hiring professionals, advocacy organizations, engineers, organizational psychologists and members of the stuttering community to gain insight for their product and create a set of guidelines to ensure employers make their procedures more equitable for people who stutter in the future. â–